Many computing devices, such as smartphones, desktops, laptops, tablets, game consoles, and the like, utilize applications which incorporate automatic speech recognition (ASR) for use over a number of different domains such as voice search and short message dictation. In order to improve the quality of speech recognition, language models (e.g., shopping, games, music, movies, etc.) are often utilized to facilitate the recognition of speech which is focused on different domains. Current drawbacks associated with the use of language models include ASR scenarios in which different domains need to be served simultaneously by a recognition service provider. In these scenarios, many potentially large language models may be required to be maintained in memory which may tax the resources of recognition service providers. It is with respect to these considerations and others that the various embodiments of the present invention have been made.